I'm planning to write code to optimize the `deepSplit` parameter in dynamicTreeCut in a way not unlike what's already been doing in `clValid`, which uses common cluster validity indexes to optimize k for `cutree`.

The problem is: dynamicTreeCut occasionally decides some genes as unclustable (i.e. cluster 0). For the calculation of validity indices, it is appropriate to also include these unassigned genes, or to remove them from the distance matrix from the input of, say, `fpc::cluster.stats()`?